Spark Connector

Overview

The Apache Gravitino Spark connector leverages the Spark DataSourceV2 interface to facilitate the management of diverse catalogs under Gravitino. This capability allows users to perform federation queries, accessing data from various catalogs through a unified interface and consistent access control.

Capabilities

1. Supports Hive catalog, Iceberg catalog, Paimon catalog, Jdbc catalog, and AWS Glue catalog.
2. Supports federation query.
3. Supports most DDL and DML SQLs.

Requirement

• Spark 3.3 or 3.4 or 3.5
• Scala 2.12 or 2.13
• JDK 8, 11 or 17

Usage

1. Build or download the package (gravitino-spark-connector-runtime-3.3, gravitino-spark-connector-runtime-3.4, gravitino-spark-connector-runtime-3.5), and place it to the classpath of Spark.
2. Configure the Spark session to use the Gravitino spark connector.

Property	Type	Default Value	Description	Required
spark.plugins	string	(none)	Gravitino spark plugin name, org.apache.gravitino.spark.connector.plugin.GravitinoSparkPlugin	Yes
spark.sql.gravitino.metalake	string	(none)	The metalake name that spark connector used to request to Gravitino.	Yes
spark.sql.gravitino.uri	string	(none)	The uri of Gravitino server address.	Yes
spark.sql.gravitino.enableIcebergSupport	string	false	Set to true to use Iceberg catalog.	No
spark.sql.gravitino.enablePaimonSupport	string	false	Set to true to use Paimon catalog.	No
spark.sql.gravitino.client.	string	(none)	The configuration key prefix for the Gravitino client config.	No

To configure the Gravitino client, use properties prefixed with spark.sql.gravitino.client.. These properties will be passed to the Gravitino client after removing the spark.sql. prefix.

Example: Setting spark.sql.gravitino.client.socketTimeoutMs is equivalent to setting gravitino.client.socketTimeoutMs for the Gravitino client.

Note: Invalid configuration properties will result in exceptions. Please see Gravitino Java client configurations for more support client configuration.

./bin/spark-sql -v \
--conf spark.plugins="org.apache.gravitino.spark.connector.plugin.GravitinoSparkPlugin" \
--conf spark.sql.gravitino.uri=http://127.0.0.1:8090 \
--conf spark.sql.gravitino.metalake=test \
--conf spark.sql.gravitino.enableIcebergSupport=true \
--conf spark.sql.gravitino.client.socketTimeoutMs=60000 \
--conf spark.sql.gravitino.client.connectionTimeoutMs=60000 \
--conf spark.sql.warehouse.dir=hdfs://127.0.0.1:9000/user/hive/warehouse-hive

3. Download corresponding runtime jars and place it to the classpath of Spark if using Iceberg catalog.

4. Execute the Spark SQL query.

Suppose there are two catalogs in the metalake test, hive for Hive catalog and iceberg for Iceberg catalog.

// use hive catalog
USE hive;
CREATE DATABASE db;
USE db;
CREATE TABLE hive_students (id INT, name STRING);
INSERT INTO hive_students VALUES (1, 'Alice'), (2, 'Bob');

// use Iceberg catalog
USE iceberg;
USE db;
CREATE TABLE IF NOT EXISTS iceberg_scores (id INT, score INT) USING iceberg;
INSERT INTO iceberg_scores VALUES (1, 95), (2, 88);

// execute federation query between hive table and iceberg table
SELECT hs.name, is.score FROM hive.db.hive_students hs JOIN iceberg_scores is ON hs.id = is.id;

info

The command SHOW CATALOGS will only display the Spark default catalog, named spark_catalog, due to limitations within the Spark catalog manager. It does not list the catalogs present in the metalake. However, after explicitly using the USE command with a specific catalog name, that catalog name then becomes visible in the output of SHOW CATALOGS.

Datatype Mapping

Gravitino spark connector support the following datatype mapping between Spark and Gravitino.

Spark Data Type	Gravitino Data Type
BooleanType	boolean
ByteType	byte
ShortType	short
IntegerType	integer
LongType	long
FloatType	float
DoubleType	double
DecimalType	decimal
StringType	string
CharType	char
VarcharType	varchar
TimestampType	timestamp with time zone
TimestampNTZType (Spark 3.4+)	timestamp without time zone
DateType	date
BinaryType	binary
ArrayType	array
MapType	map
StructType	struct

note

For Gravitino UUID type, Spark connector represents it as StringType because Spark has no native UUID type. This behavior is consistent with Spark built-in PostgreSQL JDBC mapping (uuid -> StringType).